/***************************************************************//** Removes a secondary index entry if found. @return DB_SUCCESS, DB_FAIL, or DB_OUT_OF_FILE_SPACE */ static ulint row_undo_ins_remove_sec_low( /*========================*/ ulint mode, /*!< in: BTR_MODIFY_LEAF or BTR_MODIFY_TREE, depending on whether we wish optimistic or pessimistic descent down the index tree */ dict_index_t* index, /*!< in: index */ dtuple_t* entry) /*!< in: index entry to remove */ { btr_pcur_t pcur; btr_cur_t* btr_cur; ulint err; mtr_t mtr; enum row_search_result search_result; mtr_start(&mtr); btr_cur = btr_pcur_get_btr_cur(&pcur); ut_ad(mode == BTR_MODIFY_TREE || mode == BTR_MODIFY_LEAF); search_result = row_search_index_entry(index, entry, mode, &pcur, &mtr); switch (search_result) { case ROW_NOT_FOUND: err = DB_SUCCESS; goto func_exit; case ROW_FOUND: break; case ROW_BUFFERED: case ROW_NOT_DELETED_REF: /* These are invalid outcomes, because the mode passed to row_search_index_entry() did not include any of the flags BTR_INSERT, BTR_DELETE, or BTR_DELETE_MARK. */ ut_error; } if (mode == BTR_MODIFY_LEAF) { err = btr_cur_optimistic_delete(btr_cur, &mtr) ? DB_SUCCESS : DB_FAIL; } else { ut_ad(mode == BTR_MODIFY_TREE); /* No need to distinguish RB_RECOVERY here, because we are deleting a secondary index record: the distinction between RB_NORMAL and RB_RECOVERY only matters when deleting a record that contains externally stored columns. */ ut_ad(!dict_index_is_clust(index)); btr_cur_pessimistic_delete(&err, FALSE, btr_cur, RB_NORMAL, &mtr); } func_exit: btr_pcur_close(&pcur); mtr_commit(&mtr); return(err); }
/*******************************************************************//** Skips a row reference from an undo log record. @return pointer to remaining part of undo record */ UNIV_INTERN byte* trx_undo_rec_skip_row_ref( /*======================*/ byte* ptr, /*!< in: remaining part in update undo log record, at the start of the row reference */ dict_index_t* index) /*!< in: clustered index */ { ulint ref_len; ulint i; ut_ad(index && ptr); ut_a(dict_index_is_clust(index)); ref_len = dict_index_get_n_unique(index); for (i = 0; i < ref_len; i++) { byte* field; ulint len; ulint orig_len; ptr = trx_undo_rec_get_col_val(ptr, &field, &len, &orig_len); } return(ptr); }
/***************************************************************//** Removes a secondary index entry if found. @return DB_SUCCESS, DB_FAIL, or DB_OUT_OF_FILE_SPACE */ static ulint row_undo_ins_remove_sec_low( /*========================*/ ulint mode, /*!< in: BTR_MODIFY_LEAF or BTR_MODIFY_TREE, depending on whether we wish optimistic or pessimistic descent down the index tree */ dict_index_t* index, /*!< in: index */ dtuple_t* entry) /*!< in: index entry to remove */ { btr_pcur_t pcur; btr_cur_t* btr_cur; ibool found; ibool success; ulint err; mtr_t mtr; log_free_check(); mtr_start(&mtr); found = row_search_index_entry(index, entry, mode, &pcur, &mtr); btr_cur = btr_pcur_get_btr_cur(&pcur); if (!found) { /* Not found */ btr_pcur_close(&pcur); mtr_commit(&mtr); return(DB_SUCCESS); } if (mode == BTR_MODIFY_LEAF) { success = btr_cur_optimistic_delete(btr_cur, &mtr); if (success) { err = DB_SUCCESS; } else { err = DB_FAIL; } } else { ut_ad(mode == BTR_MODIFY_TREE); /* No need to distinguish RB_RECOVERY here, because we are deleting a secondary index record: the distinction between RB_NORMAL and RB_RECOVERY only matters when deleting a record that contains externally stored columns. */ ut_ad(!dict_index_is_clust(index)); btr_cur_pessimistic_delete(&err, FALSE, btr_cur, RB_NORMAL, &mtr); } btr_pcur_close(&pcur); mtr_commit(&mtr); return(err); }
/***************************************************************//** Builds an index definition row to insert. @return DB_SUCCESS or error code */ static ulint dict_build_index_def_step( /*======================*/ que_thr_t* thr, /*!< in: query thread */ ind_node_t* node) /*!< in: index create node */ { dict_table_t* table; dict_index_t* index; dtuple_t* row; trx_t* trx; ut_ad(mutex_own(&(dict_sys->mutex))); trx = thr_get_trx(thr); index = node->index; table = dict_table_get_low(index->table_name); if (table == NULL) { return(DB_TABLE_NOT_FOUND); } trx->table_id = table->id; node->table = table; ut_ad((UT_LIST_GET_LEN(table->indexes) > 0) || dict_index_is_clust(index)); dict_hdr_get_new_id(NULL, &index->id, NULL); /* Inherit the space id from the table; we store all indexes of a table in the same tablespace */ index->space = table->space; node->page_no = FIL_NULL; row = dict_create_sys_indexes_tuple(index, node->heap); node->ind_row = row; ins_node_set_new_row(node->ind_def, row); /* Note that the index was created by this transaction. */ index->trx_id = trx->id; return(DB_SUCCESS); }
/*******************************************************************//** Builds a row reference from an undo log record. @return pointer to remaining part of undo record */ UNIV_INTERN byte* trx_undo_rec_get_row_ref( /*=====================*/ byte* ptr, /*!< in: remaining part of a copy of an undo log record, at the start of the row reference; NOTE that this copy of the undo log record must be preserved as long as the row reference is used, as we do NOT copy the data in the record! */ dict_index_t* index, /*!< in: clustered index */ dtuple_t** ref, /*!< out, own: row reference */ mem_heap_t* heap) /*!< in: memory heap from which the memory needed is allocated */ { ulint ref_len; ulint i; ut_ad(index && ptr && ref && heap); ut_a(dict_index_is_clust(index)); ref_len = dict_index_get_n_unique(index); *ref = dtuple_create(heap, ref_len); dict_index_copy_types(*ref, index, ref_len); for (i = 0; i < ref_len; i++) { dfield_t* dfield; byte* field; ulint len; ulint orig_len; dfield = dtuple_get_nth_field(*ref, i); ptr = trx_undo_rec_get_col_val(ptr, &field, &len, &orig_len); dfield_set_data(dfield, field, len); } return(ptr); }
/***********************************************************//** Determines if it is possible to remove a secondary index entry. Removal is possible if the secondary index entry does not refer to any not delete marked version of a clustered index record where DB_TRX_ID is newer than the purge view. NOTE: This function should only be called by the purge thread, only while holding a latch on the leaf page of the secondary index entry (or keeping the buffer pool watch on the page). It is possible that this function first returns TRUE and then FALSE, if a user transaction inserts a record that the secondary index entry would refer to. However, in that case, the user transaction would also re-insert the secondary index entry after purge has removed it and released the leaf page latch. @return TRUE if the secondary index record can be purged */ UNIV_INTERN ibool row_purge_poss_sec( /*===============*/ purge_node_t* node, /*!< in/out: row purge node */ dict_index_t* index, /*!< in: secondary index */ const dtuple_t* entry) /*!< in: secondary index entry */ { ibool can_delete; mtr_t mtr; ut_ad(!dict_index_is_clust(index)); mtr_start(&mtr); can_delete = !row_purge_reposition_pcur(BTR_SEARCH_LEAF, node, &mtr) || !row_vers_old_has_index_entry(TRUE, btr_pcur_get_rec(&node->pcur), &mtr, index, entry); btr_pcur_commit_specify_mtr(&node->pcur, &mtr); return(can_delete); }
/*********************************************************************//** Fetches the clustered index record for a secondary index record. The latches on the secondary index record are preserved. @return record or NULL, if no record found */ UNIV_INTERN rec_t* row_get_clust_rec( /*==============*/ ulint mode, /*!< in: BTR_MODIFY_LEAF, ... */ const rec_t* rec, /*!< in: record in a secondary index */ dict_index_t* index, /*!< in: secondary index */ dict_index_t** clust_index,/*!< out: clustered index */ mtr_t* mtr) /*!< in: mtr */ { mem_heap_t* heap; dtuple_t* ref; dict_table_t* table; btr_pcur_t pcur; ibool found; rec_t* clust_rec; ut_ad(!dict_index_is_clust(index)); table = index->table; heap = mem_heap_create(256); ref = row_build_row_ref(ROW_COPY_POINTERS, index, rec, heap); found = row_search_on_row_ref(&pcur, mode, table, ref, mtr); clust_rec = found ? btr_pcur_get_rec(&pcur) : NULL; mem_heap_free(heap); btr_pcur_close(&pcur); *clust_index = dict_table_get_first_index(table); return(clust_rec); }
/**********************************************************************//** Reports in the undo log of an insert of a clustered index record. @return offset of the inserted entry on the page if succeed, 0 if fail */ static ulint trx_undo_page_report_insert( /*========================*/ page_t* undo_page, /*!< in: undo log page */ trx_t* trx, /*!< in: transaction */ dict_index_t* index, /*!< in: clustered index */ const dtuple_t* clust_entry, /*!< in: index entry which will be inserted to the clustered index */ mtr_t* mtr) /*!< in: mtr */ { ulint first_free; byte* ptr; ulint i; ut_ad(dict_index_is_clust(index)); ut_ad(mach_read_from_2(undo_page + TRX_UNDO_PAGE_HDR + TRX_UNDO_PAGE_TYPE) == TRX_UNDO_INSERT); first_free = mach_read_from_2(undo_page + TRX_UNDO_PAGE_HDR + TRX_UNDO_PAGE_FREE); ptr = undo_page + first_free; ut_ad(first_free <= UNIV_PAGE_SIZE); if (trx_undo_left(undo_page, ptr) < 2 + 1 + 11 + 11) { /* Not enough space for writing the general parameters */ return(0); } /* Reserve 2 bytes for the pointer to the next undo log record */ ptr += 2; /* Store first some general parameters to the undo log */ *ptr++ = TRX_UNDO_INSERT_REC; ptr += mach_dulint_write_much_compressed(ptr, trx->undo_no); ptr += mach_dulint_write_much_compressed(ptr, index->table->id); /*----------------------------------------*/ /* Store then the fields required to uniquely determine the record to be inserted in the clustered index */ for (i = 0; i < dict_index_get_n_unique(index); i++) { const dfield_t* field = dtuple_get_nth_field(clust_entry, i); ulint flen = dfield_get_len(field); if (trx_undo_left(undo_page, ptr) < 5) { return(0); } ptr += mach_write_compressed(ptr, flen); if (flen != UNIV_SQL_NULL) { if (trx_undo_left(undo_page, ptr) < flen) { return(0); } ut_memcpy(ptr, dfield_get_data(field), flen); ptr += flen; } } return(trx_undo_page_set_next_prev_and_add(undo_page, ptr, mtr)); }
/*******************************************************************//** Builds a partial row from an update undo log record. It contains the columns which occur as ordering in any index of the table. @return pointer to remaining part of undo record */ UNIV_INTERN byte* trx_undo_rec_get_partial_row( /*=========================*/ byte* ptr, /*!< in: remaining part in update undo log record of a suitable type, at the start of the stored index columns; NOTE that this copy of the undo log record must be preserved as long as the partial row is used, as we do NOT copy the data in the record! */ dict_index_t* index, /*!< in: clustered index */ dtuple_t** row, /*!< out, own: partial row */ ibool ignore_prefix, /*!< in: flag to indicate if we expect blob prefixes in undo. Used only in the assertion. */ mem_heap_t* heap) /*!< in: memory heap from which the memory needed is allocated */ { const byte* end_ptr; ulint row_len; ut_ad(index); ut_ad(ptr); ut_ad(row); ut_ad(heap); ut_ad(dict_index_is_clust(index)); row_len = dict_table_get_n_cols(index->table); *row = dtuple_create(heap, row_len); dict_table_copy_types(*row, index->table); end_ptr = ptr + mach_read_from_2(ptr); ptr += 2; while (ptr != end_ptr) { dfield_t* dfield; byte* field; ulint field_no; const dict_col_t* col; ulint col_no; ulint len; ulint orig_len; ptr = trx_undo_update_rec_get_field_no(ptr, &field_no); col = dict_index_get_nth_col(index, field_no); col_no = dict_col_get_no(col); ptr = trx_undo_rec_get_col_val(ptr, &field, &len, &orig_len); dfield = dtuple_get_nth_field(*row, col_no); dfield_set_data(dfield, field, len); if (len != UNIV_SQL_NULL && len >= UNIV_EXTERN_STORAGE_FIELD) { dfield_set_len(dfield, len - UNIV_EXTERN_STORAGE_FIELD); dfield_set_ext(dfield); /* If the prefix of this column is indexed, ensure that enough prefix is stored in the undo log record. */ if (!ignore_prefix && col->ord_part) { ut_a(dfield_get_len(dfield) >= 2 * BTR_EXTERN_FIELD_REF_SIZE); ut_a(dict_table_get_format(index->table) >= DICT_TF_FORMAT_ZIP || dfield_get_len(dfield) >= REC_MAX_INDEX_COL_LEN + BTR_EXTERN_FIELD_REF_SIZE); } } } return(ptr); }
/**************************************************************//** Moves parts of long fields in entry to the big record vector so that the size of tuple drops below the maximum record size allowed in the database. Moves data only from those fields which are not necessary to determine uniquely the insertion place of the tuple in the index. @return own: created big record vector, NULL if we are not able to shorten the entry enough, i.e., if there are too many fixed-length or short fields in entry or the index is clustered */ UNIV_INTERN big_rec_t* dtuple_convert_big_rec( /*===================*/ dict_index_t* index, /*!< in: index */ dtuple_t* entry, /*!< in/out: index entry */ ulint* n_ext) /*!< in/out: number of externally stored columns */ { mem_heap_t* heap; big_rec_t* vector; dfield_t* dfield; dict_field_t* ifield; ulint size; ulint n_fields; ulint local_len; ulint local_prefix_len; if (UNIV_UNLIKELY(!dict_index_is_clust(index))) { return(NULL); } if (dict_table_get_format(index->table) < DICT_TF_FORMAT_ZIP) { /* up to MySQL 5.1: store a 768-byte prefix locally */ local_len = BTR_EXTERN_FIELD_REF_SIZE + DICT_MAX_INDEX_COL_LEN; } else { /* new-format table: do not store any BLOB prefix locally */ local_len = BTR_EXTERN_FIELD_REF_SIZE; } ut_a(dtuple_check_typed_no_assert(entry)); size = rec_get_converted_size(index, entry, *n_ext); if (UNIV_UNLIKELY(size > 1000000000)) { fprintf(stderr, "InnoDB: Warning: tuple size very big: %lu\n", (ulong) size); fputs("InnoDB: Tuple contents: ", stderr); dtuple_print(stderr, entry); putc('\n', stderr); } heap = mem_heap_create(size + dtuple_get_n_fields(entry) * sizeof(big_rec_field_t) + 1000); vector = mem_heap_alloc(heap, sizeof(big_rec_t)); vector->heap = heap; vector->fields = mem_heap_alloc(heap, dtuple_get_n_fields(entry) * sizeof(big_rec_field_t)); /* Decide which fields to shorten: the algorithm is to look for a variable-length field that yields the biggest savings when stored externally */ n_fields = 0; while (page_zip_rec_needs_ext(rec_get_converted_size(index, entry, *n_ext), dict_table_is_comp(index->table), dict_index_get_n_fields(index), dict_table_zip_size(index->table))) { ulint i; ulint longest = 0; ulint longest_i = ULINT_MAX; byte* data; big_rec_field_t* b; for (i = dict_index_get_n_unique_in_tree(index); i < dtuple_get_n_fields(entry); i++) { ulint savings; dfield = dtuple_get_nth_field(entry, i); ifield = dict_index_get_nth_field(index, i); /* Skip fixed-length, NULL, externally stored, or short columns */ if (ifield->fixed_len || dfield_is_null(dfield) || dfield_is_ext(dfield) || dfield_get_len(dfield) <= local_len || dfield_get_len(dfield) <= BTR_EXTERN_FIELD_REF_SIZE * 2) { goto skip_field; } savings = dfield_get_len(dfield) - local_len; /* Check that there would be savings */ if (longest >= savings) { goto skip_field; } longest_i = i; longest = savings; skip_field: continue; } if (!longest) { /* Cannot shorten more */ mem_heap_free(heap); return(NULL); } /* Move data from field longest_i to big rec vector. We store the first bytes locally to the record. Then we can calculate all ordering fields in all indexes from locally stored data. */ dfield = dtuple_get_nth_field(entry, longest_i); ifield = dict_index_get_nth_field(index, longest_i); local_prefix_len = local_len - BTR_EXTERN_FIELD_REF_SIZE; b = &vector->fields[n_fields]; b->field_no = longest_i; b->len = dfield_get_len(dfield) - local_prefix_len; b->data = (char*) dfield_get_data(dfield) + local_prefix_len; /* Allocate the locally stored part of the column. */ data = mem_heap_alloc(heap, local_len); /* Copy the local prefix. */ memcpy(data, dfield_get_data(dfield), local_prefix_len); /* Clear the extern field reference (BLOB pointer). */ memset(data + local_prefix_len, 0, BTR_EXTERN_FIELD_REF_SIZE); #if 0 /* The following would fail the Valgrind checks in page_cur_insert_rec_low() and page_cur_insert_rec_zip(). The BLOB pointers in the record will be initialized after the record and the BLOBs have been written. */ UNIV_MEM_ALLOC(data + local_prefix_len, BTR_EXTERN_FIELD_REF_SIZE); #endif dfield_set_data(dfield, data, local_len); dfield_set_ext(dfield); n_fields++; (*n_ext)++; ut_ad(n_fields < dtuple_get_n_fields(entry)); } vector->n_fields = n_fields; return(vector); }
/**********************************************************************//** Reports in the undo log of an update or delete marking of a clustered index record. @return byte offset of the inserted undo log entry on the page if succeed, 0 if fail */ static ulint trx_undo_page_report_modify( /*========================*/ page_t* undo_page, /*!< in: undo log page */ trx_t* trx, /*!< in: transaction */ dict_index_t* index, /*!< in: clustered index where update or delete marking is done */ const rec_t* rec, /*!< in: clustered index record which has NOT yet been modified */ const ulint* offsets, /*!< in: rec_get_offsets(rec, index) */ const upd_t* update, /*!< in: update vector which tells the columns to be updated; in the case of a delete, this should be set to NULL */ ulint cmpl_info, /*!< in: compiler info on secondary index updates */ mtr_t* mtr) /*!< in: mtr */ { dict_table_t* table; ulint first_free; byte* ptr; const byte* field; ulint flen; ulint col_no; ulint type_cmpl; byte* type_cmpl_ptr; ulint i; trx_id_t trx_id; ibool ignore_prefix = FALSE; byte ext_buf[REC_MAX_INDEX_COL_LEN + BTR_EXTERN_FIELD_REF_SIZE]; ut_a(dict_index_is_clust(index)); ut_ad(rec_offs_validate(rec, index, offsets)); ut_ad(mach_read_from_2(undo_page + TRX_UNDO_PAGE_HDR + TRX_UNDO_PAGE_TYPE) == TRX_UNDO_UPDATE); table = index->table; first_free = mach_read_from_2(undo_page + TRX_UNDO_PAGE_HDR + TRX_UNDO_PAGE_FREE); ptr = undo_page + first_free; ut_ad(first_free <= UNIV_PAGE_SIZE); if (trx_undo_left(undo_page, ptr) < 50) { /* NOTE: the value 50 must be big enough so that the general fields written below fit on the undo log page */ return(0); } /* Reserve 2 bytes for the pointer to the next undo log record */ ptr += 2; /* Store first some general parameters to the undo log */ if (!update) { type_cmpl = TRX_UNDO_DEL_MARK_REC; } else if (rec_get_deleted_flag(rec, dict_table_is_comp(table))) { type_cmpl = TRX_UNDO_UPD_DEL_REC; /* We are about to update a delete marked record. We don't typically need the prefix in this case unless the delete marking is done by the same transaction (which we check below). */ ignore_prefix = TRUE; } else { type_cmpl = TRX_UNDO_UPD_EXIST_REC; } type_cmpl |= cmpl_info * TRX_UNDO_CMPL_INFO_MULT; type_cmpl_ptr = ptr; *ptr++ = (byte) type_cmpl; ptr += mach_dulint_write_much_compressed(ptr, trx->undo_no); ptr += mach_dulint_write_much_compressed(ptr, table->id); /*----------------------------------------*/ /* Store the state of the info bits */ *ptr++ = (byte) rec_get_info_bits(rec, dict_table_is_comp(table)); /* Store the values of the system columns */ field = rec_get_nth_field(rec, offsets, dict_index_get_sys_col_pos( index, DATA_TRX_ID), &flen); ut_ad(flen == DATA_TRX_ID_LEN); trx_id = trx_read_trx_id(field); /* If it is an update of a delete marked record, then we are allowed to ignore blob prefixes if the delete marking was done by some other trx as it must have committed by now for us to allow an over-write. */ if (ignore_prefix) { ignore_prefix = ut_dulint_cmp(trx_id, trx->id) != 0; } ptr += mach_dulint_write_compressed(ptr, trx_id); field = rec_get_nth_field(rec, offsets, dict_index_get_sys_col_pos( index, DATA_ROLL_PTR), &flen); ut_ad(flen == DATA_ROLL_PTR_LEN); ptr += mach_dulint_write_compressed(ptr, trx_read_roll_ptr(field)); /*----------------------------------------*/ /* Store then the fields required to uniquely determine the record which will be modified in the clustered index */ for (i = 0; i < dict_index_get_n_unique(index); i++) { field = rec_get_nth_field(rec, offsets, i, &flen); /* The ordering columns must not be stored externally. */ ut_ad(!rec_offs_nth_extern(offsets, i)); ut_ad(dict_index_get_nth_col(index, i)->ord_part); if (trx_undo_left(undo_page, ptr) < 5) { return(0); } ptr += mach_write_compressed(ptr, flen); if (flen != UNIV_SQL_NULL) { if (trx_undo_left(undo_page, ptr) < flen) { return(0); } ut_memcpy(ptr, field, flen); ptr += flen; } } /*----------------------------------------*/ /* Save to the undo log the old values of the columns to be updated. */ if (update) { if (trx_undo_left(undo_page, ptr) < 5) { return(0); } ptr += mach_write_compressed(ptr, upd_get_n_fields(update)); for (i = 0; i < upd_get_n_fields(update); i++) { ulint pos = upd_get_nth_field(update, i)->field_no; /* Write field number to undo log */ if (trx_undo_left(undo_page, ptr) < 5) { return(0); } ptr += mach_write_compressed(ptr, pos); /* Save the old value of field */ field = rec_get_nth_field(rec, offsets, pos, &flen); if (trx_undo_left(undo_page, ptr) < 15) { return(0); } if (rec_offs_nth_extern(offsets, pos)) { ptr = trx_undo_page_report_modify_ext( ptr, dict_index_get_nth_col(index, pos) ->ord_part && !ignore_prefix && flen < REC_MAX_INDEX_COL_LEN ? ext_buf : NULL, dict_table_zip_size(table), &field, &flen); /* Notify purge that it eventually has to free the old externally stored field */ trx->update_undo->del_marks = TRUE; *type_cmpl_ptr |= TRX_UNDO_UPD_EXTERN; } else { ptr += mach_write_compressed(ptr, flen); } if (flen != UNIV_SQL_NULL) { if (trx_undo_left(undo_page, ptr) < flen) { return(0); } ut_memcpy(ptr, field, flen); ptr += flen; } } } /*----------------------------------------*/ /* In the case of a delete marking, and also in the case of an update where any ordering field of any index changes, store the values of all columns which occur as ordering fields in any index. This info is used in the purge of old versions where we use it to build and search the delete marked index records, to look if we can remove them from the index tree. Note that starting from 4.0.14 also externally stored fields can be ordering in some index. Starting from 5.2, we no longer store REC_MAX_INDEX_COL_LEN first bytes to the undo log record, but we can construct the column prefix fields in the index by fetching the first page of the BLOB that is pointed to by the clustered index. This works also in crash recovery, because all pages (including BLOBs) are recovered before anything is rolled back. */ if (!update || !(cmpl_info & UPD_NODE_NO_ORD_CHANGE)) { byte* old_ptr = ptr; trx->update_undo->del_marks = TRUE; if (trx_undo_left(undo_page, ptr) < 5) { return(0); } /* Reserve 2 bytes to write the number of bytes the stored fields take in this undo record */ ptr += 2; for (col_no = 0; col_no < dict_table_get_n_cols(table); col_no++) { const dict_col_t* col = dict_table_get_nth_col(table, col_no); if (col->ord_part) { ulint pos; /* Write field number to undo log */ if (trx_undo_left(undo_page, ptr) < 5 + 15) { return(0); } pos = dict_index_get_nth_col_pos(index, col_no); ptr += mach_write_compressed(ptr, pos); /* Save the old value of field */ field = rec_get_nth_field(rec, offsets, pos, &flen); if (rec_offs_nth_extern(offsets, pos)) { ptr = trx_undo_page_report_modify_ext( ptr, flen < REC_MAX_INDEX_COL_LEN && !ignore_prefix ? ext_buf : NULL, dict_table_zip_size(table), &field, &flen); } else { ptr += mach_write_compressed( ptr, flen); } if (flen != UNIV_SQL_NULL) { if (trx_undo_left(undo_page, ptr) < flen) { return(0); } ut_memcpy(ptr, field, flen); ptr += flen; } } } mach_write_to_2(old_ptr, ptr - old_ptr); } /*----------------------------------------*/ /* Write pointers to the previous and the next undo log records */ if (trx_undo_left(undo_page, ptr) < 2) { return(0); } mach_write_to_2(ptr, first_free); ptr += 2; mach_write_to_2(undo_page + first_free, ptr - undo_page); mach_write_to_2(undo_page + TRX_UNDO_PAGE_HDR + TRX_UNDO_PAGE_FREE, ptr - undo_page); /* Write to the REDO log about this change in the UNDO log */ trx_undof_page_add_undo_rec_log(undo_page, first_free, ptr - undo_page, mtr); return(first_free); }
/*******************************************************************//** Builds from a secondary index record a row reference with which we can search the clustered index record. @return own: row reference built; see the NOTE below! */ UNIV_INTERN dtuple_t* row_build_row_ref( /*==============*/ ulint type, /*!< in: ROW_COPY_DATA, or ROW_COPY_POINTERS: the former copies also the data fields to heap, whereas the latter only places pointers to data fields on the index page */ dict_index_t* index, /*!< in: secondary index */ const rec_t* rec, /*!< in: record in the index; NOTE: in the case ROW_COPY_POINTERS the data fields in the row will point directly into this record, therefore, the buffer page of this record must be at least s-latched and the latch held as long as the row reference is used! */ mem_heap_t* heap) /*!< in: memory heap from which the memory needed is allocated */ { dict_table_t* table; dict_index_t* clust_index; dfield_t* dfield; dtuple_t* ref; const byte* field; ulint len; ulint ref_len; ulint pos; byte* buf; ulint clust_col_prefix_len; ulint i; mem_heap_t* tmp_heap = NULL; ulint offsets_[REC_OFFS_NORMAL_SIZE]; ulint* offsets = offsets_; rec_offs_init(offsets_); ut_ad(index && rec && heap); ut_ad(!dict_index_is_clust(index)); offsets = rec_get_offsets(rec, index, offsets, ULINT_UNDEFINED, &tmp_heap); /* Secondary indexes must not contain externally stored columns. */ ut_ad(!rec_offs_any_extern(offsets)); if (type == ROW_COPY_DATA) { /* Take a copy of rec to heap */ buf = mem_heap_alloc(heap, rec_offs_size(offsets)); rec = rec_copy(buf, rec, offsets); /* Avoid a debug assertion in rec_offs_validate(). */ rec_offs_make_valid(rec, index, offsets); } table = index->table; clust_index = dict_table_get_first_index(table); ref_len = dict_index_get_n_unique(clust_index); ref = dtuple_create(heap, ref_len); dict_index_copy_types(ref, clust_index, ref_len); for (i = 0; i < ref_len; i++) { dfield = dtuple_get_nth_field(ref, i); pos = dict_index_get_nth_field_pos(index, clust_index, i); ut_a(pos != ULINT_UNDEFINED); field = rec_get_nth_field(rec, offsets, pos, &len); dfield_set_data(dfield, field, len); /* If the primary key contains a column prefix, then the secondary index may contain a longer prefix of the same column, or the full column, and we must adjust the length accordingly. */ clust_col_prefix_len = dict_index_get_nth_field( clust_index, i)->prefix_len; if (clust_col_prefix_len > 0) { if (len != UNIV_SQL_NULL) { const dtype_t* dtype = dfield_get_type(dfield); dfield_set_len(dfield, dtype_get_at_most_n_mbchars( dtype->prtype, dtype->mbminlen, dtype->mbmaxlen, clust_col_prefix_len, len, (char*) field)); } } } ut_ad(dtuple_check_typed(ref)); if (tmp_heap) { mem_heap_free(tmp_heap); } return(ref); }
/*******************************************************************//** An inverse function to row_build_index_entry. Builds a row from a record in a clustered index. @return own: row built; see the NOTE below! */ UNIV_INTERN dtuple_t* row_build( /*======*/ ulint type, /*!< in: ROW_COPY_POINTERS or ROW_COPY_DATA; the latter copies also the data fields to heap while the first only places pointers to data fields on the index page, and thus is more efficient */ const dict_index_t* index, /*!< in: clustered index */ const rec_t* rec, /*!< in: record in the clustered index; NOTE: in the case ROW_COPY_POINTERS the data fields in the row will point directly into this record, therefore, the buffer page of this record must be at least s-latched and the latch held as long as the row dtuple is used! */ const ulint* offsets,/*!< in: rec_get_offsets(rec,index) or NULL, in which case this function will invoke rec_get_offsets() */ const dict_table_t* col_table, /*!< in: table, to check which externally stored columns occur in the ordering columns of an index, or NULL if index->table should be consulted instead */ row_ext_t** ext, /*!< out, own: cache of externally stored column prefixes, or NULL */ mem_heap_t* heap) /*!< in: memory heap from which the memory needed is allocated */ { dtuple_t* row; const dict_table_t* table; ulint n_fields; ulint n_ext_cols; ulint* ext_cols = NULL; /* remove warning */ ulint len; ulint row_len; byte* buf; ulint i; ulint j; mem_heap_t* tmp_heap = NULL; ulint offsets_[REC_OFFS_NORMAL_SIZE]; rec_offs_init(offsets_); ut_ad(index && rec && heap); ut_ad(dict_index_is_clust(index)); ut_ad(!mutex_own(&kernel_mutex)); if (!offsets) { offsets = rec_get_offsets(rec, index, offsets_, ULINT_UNDEFINED, &tmp_heap); } else { ut_ad(rec_offs_validate(rec, index, offsets)); } #if defined UNIV_DEBUG || defined UNIV_BLOB_LIGHT_DEBUG if (rec_offs_any_null_extern(rec, offsets)) { /* This condition can occur during crash recovery before trx_rollback_active() has completed execution, or when a concurrently executing row_ins_index_entry_low() has committed the B-tree mini-transaction but has not yet managed to restore the cursor position for writing the big_rec. */ ut_a(trx_undo_roll_ptr_is_insert( row_get_rec_roll_ptr(rec, index, offsets))); } #endif /* UNIV_DEBUG || UNIV_BLOB_LIGHT_DEBUG */ if (type != ROW_COPY_POINTERS) { /* Take a copy of rec to heap */ buf = mem_heap_alloc(heap, rec_offs_size(offsets)); rec = rec_copy(buf, rec, offsets); /* Avoid a debug assertion in rec_offs_validate(). */ rec_offs_make_valid(rec, index, (ulint*) offsets); } table = index->table; row_len = dict_table_get_n_cols(table); row = dtuple_create(heap, row_len); dict_table_copy_types(row, table); dtuple_set_info_bits(row, rec_get_info_bits( rec, dict_table_is_comp(table))); n_fields = rec_offs_n_fields(offsets); n_ext_cols = rec_offs_n_extern(offsets); if (n_ext_cols) { ext_cols = mem_heap_alloc(heap, n_ext_cols * sizeof *ext_cols); } for (i = j = 0; i < n_fields; i++) { dict_field_t* ind_field = dict_index_get_nth_field(index, i); const dict_col_t* col = dict_field_get_col(ind_field); ulint col_no = dict_col_get_no(col); dfield_t* dfield = dtuple_get_nth_field(row, col_no); if (ind_field->prefix_len == 0) { const byte* field = rec_get_nth_field( rec, offsets, i, &len); dfield_set_data(dfield, field, len); } if (rec_offs_nth_extern(offsets, i)) { dfield_set_ext(dfield); if (UNIV_LIKELY_NULL(col_table)) { ut_a(col_no < dict_table_get_n_cols(col_table)); col = dict_table_get_nth_col( col_table, col_no); } if (col->ord_part) { /* We will have to fetch prefixes of externally stored columns that are referenced by column prefixes. */ ext_cols[j++] = col_no; } } } ut_ad(dtuple_check_typed(row)); if (!ext) { /* REDUNDANT and COMPACT formats store a local 768-byte prefix of each externally stored column. No cache is needed. */ ut_ad(dict_table_get_format(index->table) < DICT_TF_FORMAT_ZIP); } else if (j) { *ext = row_ext_create(j, ext_cols, row, dict_table_zip_size(index->table), heap); } else { *ext = NULL; } if (tmp_heap) { mem_heap_free(tmp_heap); } return(row); }
/*****************************************************************//** Constructs the last committed version of a clustered index record, which should be seen by a semi-consistent read. @return DB_SUCCESS or DB_MISSING_HISTORY */ UNIV_INTERN ulint row_vers_build_for_semi_consistent_read( /*====================================*/ const rec_t* rec, /*!< in: record in a clustered index; the caller must have a latch on the page; this latch locks the top of the stack of versions of this records */ mtr_t* mtr, /*!< in: mtr holding the latch on rec */ dict_index_t* index, /*!< in: the clustered index */ ulint** offsets,/*!< in/out: offsets returned by rec_get_offsets(rec, index) */ mem_heap_t** offset_heap,/*!< in/out: memory heap from which the offsets are allocated */ mem_heap_t* in_heap,/*!< in: memory heap from which the memory for *old_vers is allocated; memory for possible intermediate versions is allocated and freed locally within the function */ const rec_t** old_vers)/*!< out: rec, old version, or NULL if the record does not exist in the view, that is, it was freshly inserted afterwards */ { const rec_t* version; mem_heap_t* heap = NULL; byte* buf; ulint err; trx_id_t rec_trx_id = ut_dulint_zero; ut_ad(dict_index_is_clust(index)); ut_ad(mtr_memo_contains_page(mtr, rec, MTR_MEMO_PAGE_X_FIX) || mtr_memo_contains_page(mtr, rec, MTR_MEMO_PAGE_S_FIX)); #ifdef UNIV_SYNC_DEBUG ut_ad(!rw_lock_own(&(purge_sys->latch), RW_LOCK_SHARED)); #endif /* UNIV_SYNC_DEBUG */ ut_ad(rec_offs_validate(rec, index, *offsets)); rw_lock_s_lock(&(purge_sys->latch)); /* The S-latch on purge_sys prevents the purge view from changing. Thus, if we have an uncommitted transaction at this point, then purge cannot remove its undo log even if the transaction could commit now. */ version = rec; for (;;) { trx_t* version_trx; mem_heap_t* heap2; rec_t* prev_version; trx_id_t version_trx_id; version_trx_id = row_get_rec_trx_id(version, index, *offsets); if (rec == version) { rec_trx_id = version_trx_id; } mutex_enter(&kernel_mutex); version_trx = trx_get_on_id(version_trx_id); if (version_trx && (version_trx->conc_state == TRX_COMMITTED_IN_MEMORY || version_trx->conc_state == TRX_NOT_STARTED)) { version_trx = NULL; } mutex_exit(&kernel_mutex); if (!version_trx) { /* We found a version that belongs to a committed transaction: return it. */ #if defined UNIV_DEBUG || defined UNIV_BLOB_LIGHT_DEBUG ut_a(!rec_offs_any_null_extern(version, *offsets)); #endif /* UNIV_DEBUG || UNIV_BLOB_LIGHT_DEBUG */ if (rec == version) { *old_vers = rec; err = DB_SUCCESS; break; } /* We assume that a rolled-back transaction stays in TRX_ACTIVE state until all the changes have been rolled back and the transaction is removed from the global list of transactions. */ if (!ut_dulint_cmp(rec_trx_id, version_trx_id)) { /* The transaction was committed while we searched for earlier versions. Return the current version as a semi-consistent read. */ version = rec; *offsets = rec_get_offsets(version, index, *offsets, ULINT_UNDEFINED, offset_heap); } buf = mem_heap_alloc(in_heap, rec_offs_size(*offsets)); *old_vers = rec_copy(buf, version, *offsets); rec_offs_make_valid(*old_vers, index, *offsets); err = DB_SUCCESS; break; } heap2 = heap; heap = mem_heap_create(1024); err = trx_undo_prev_version_build(rec, mtr, version, index, *offsets, heap, &prev_version); if (heap2) { mem_heap_free(heap2); /* free version */ } if (UNIV_UNLIKELY(err != DB_SUCCESS)) { break; } if (prev_version == NULL) { /* It was a freshly inserted version */ *old_vers = NULL; err = DB_SUCCESS; break; } version = prev_version; *offsets = rec_get_offsets(version, index, *offsets, ULINT_UNDEFINED, offset_heap); #if defined UNIV_DEBUG || defined UNIV_BLOB_LIGHT_DEBUG ut_a(!rec_offs_any_null_extern(version, *offsets)); #endif /* UNIV_DEBUG || UNIV_BLOB_LIGHT_DEBUG */ }/* for (;;) */ if (heap) { mem_heap_free(heap); } rw_lock_s_unlock(&(purge_sys->latch)); return(err); }
/*****************************************************************//** Constructs the version of a clustered index record which a consistent read should see. We assume that the trx id stored in rec is such that the consistent read should not see rec in its present version. @return DB_SUCCESS or DB_MISSING_HISTORY */ UNIV_INTERN ulint row_vers_build_for_consistent_read( /*===============================*/ const rec_t* rec, /*!< in: record in a clustered index; the caller must have a latch on the page; this latch locks the top of the stack of versions of this records */ mtr_t* mtr, /*!< in: mtr holding the latch on rec */ dict_index_t* index, /*!< in: the clustered index */ ulint** offsets,/*!< in/out: offsets returned by rec_get_offsets(rec, index) */ read_view_t* view, /*!< in: the consistent read view */ mem_heap_t** offset_heap,/*!< in/out: memory heap from which the offsets are allocated */ mem_heap_t* in_heap,/*!< in: memory heap from which the memory for *old_vers is allocated; memory for possible intermediate versions is allocated and freed locally within the function */ rec_t** old_vers)/*!< out, own: old version, or NULL if the record does not exist in the view, that is, it was freshly inserted afterwards */ { const rec_t* version; rec_t* prev_version; trx_id_t trx_id; mem_heap_t* heap = NULL; byte* buf; ulint err; ut_ad(dict_index_is_clust(index)); ut_ad(mtr_memo_contains_page(mtr, rec, MTR_MEMO_PAGE_X_FIX) || mtr_memo_contains_page(mtr, rec, MTR_MEMO_PAGE_S_FIX)); #ifdef UNIV_SYNC_DEBUG ut_ad(!rw_lock_own(&(purge_sys->latch), RW_LOCK_SHARED)); #endif /* UNIV_SYNC_DEBUG */ ut_ad(rec_offs_validate(rec, index, *offsets)); trx_id = row_get_rec_trx_id(rec, index, *offsets); ut_ad(!read_view_sees_trx_id(view, trx_id)); rw_lock_s_lock(&(purge_sys->latch)); version = rec; for (;;) { mem_heap_t* heap2 = heap; trx_undo_rec_t* undo_rec; roll_ptr_t roll_ptr; undo_no_t undo_no; heap = mem_heap_create(1024); /* If we have high-granularity consistent read view and creating transaction of the view is the same as trx_id in the record we see this record only in the case when undo_no of the record is < undo_no in the view. */ if (view->type == VIEW_HIGH_GRANULARITY && ut_dulint_cmp(view->creator_trx_id, trx_id) == 0) { roll_ptr = row_get_rec_roll_ptr(version, index, *offsets); undo_rec = trx_undo_get_undo_rec_low(roll_ptr, heap); undo_no = trx_undo_rec_get_undo_no(undo_rec); mem_heap_empty(heap); if (ut_dulint_cmp(view->undo_no, undo_no) > 0) { /* The view already sees this version: we can copy it to in_heap and return */ #if defined UNIV_DEBUG || defined UNIV_BLOB_LIGHT_DEBUG ut_a(!rec_offs_any_null_extern( version, *offsets)); #endif /* UNIV_DEBUG || UNIV_BLOB_LIGHT_DEBUG */ buf = mem_heap_alloc(in_heap, rec_offs_size(*offsets)); *old_vers = rec_copy(buf, version, *offsets); rec_offs_make_valid(*old_vers, index, *offsets); err = DB_SUCCESS; break; } } err = trx_undo_prev_version_build(rec, mtr, version, index, *offsets, heap, &prev_version); if (heap2) { mem_heap_free(heap2); /* free version */ } if (err != DB_SUCCESS) { break; } if (prev_version == NULL) { /* It was a freshly inserted version */ *old_vers = NULL; err = DB_SUCCESS; break; } *offsets = rec_get_offsets(prev_version, index, *offsets, ULINT_UNDEFINED, offset_heap); #if defined UNIV_DEBUG || defined UNIV_BLOB_LIGHT_DEBUG ut_a(!rec_offs_any_null_extern(prev_version, *offsets)); #endif /* UNIV_DEBUG || UNIV_BLOB_LIGHT_DEBUG */ trx_id = row_get_rec_trx_id(prev_version, index, *offsets); if (read_view_sees_trx_id(view, trx_id)) { /* The view already sees this version: we can copy it to in_heap and return */ buf = mem_heap_alloc(in_heap, rec_offs_size(*offsets)); *old_vers = rec_copy(buf, prev_version, *offsets); rec_offs_make_valid(*old_vers, index, *offsets); err = DB_SUCCESS; break; } version = prev_version; }/* for (;;) */ mem_heap_free(heap); rw_lock_s_unlock(&(purge_sys->latch)); return(err); }
/*****************************************************************//** When an insert or purge to a table is performed, this function builds the entry to be inserted into or purged from an index on the table. @return index entry which should be inserted or purged, or NULL if the externally stored columns in the clustered index record are unavailable and ext != NULL */ UNIV_INTERN dtuple_t* row_build_index_entry( /*==================*/ const dtuple_t* row, /*!< in: row which should be inserted or purged */ row_ext_t* ext, /*!< in: externally stored column prefixes, or NULL */ dict_index_t* index, /*!< in: index on the table */ mem_heap_t* heap) /*!< in: memory heap from which the memory for the index entry is allocated */ { dtuple_t* entry; ulint entry_len; ulint i; ut_ad(row && index && heap); ut_ad(dtuple_check_typed(row)); entry_len = dict_index_get_n_fields(index); entry = dtuple_create(heap, entry_len); if (UNIV_UNLIKELY(index->type & DICT_UNIVERSAL)) { dtuple_set_n_fields_cmp(entry, entry_len); /* There may only be externally stored columns in a clustered index B-tree of a user table. */ ut_a(!ext); } else { dtuple_set_n_fields_cmp( entry, dict_index_get_n_unique_in_tree(index)); } for (i = 0; i < entry_len; i++) { const dict_field_t* ind_field = dict_index_get_nth_field(index, i); const dict_col_t* col = ind_field->col; ulint col_no = dict_col_get_no(col); dfield_t* dfield = dtuple_get_nth_field(entry, i); const dfield_t* dfield2 = dtuple_get_nth_field(row, col_no); ulint len = dfield_get_len(dfield2); dfield_copy(dfield, dfield2); if (dfield_is_null(dfield)) { continue; } if (ind_field->prefix_len == 0 && (!dfield_is_ext(dfield) || dict_index_is_clust(index))) { /* The dfield_copy() above suffices for columns that are stored in-page, or for clustered index record columns that are not part of a column prefix in the PRIMARY KEY. */ continue; } /* If the column is stored externally (off-page) in the clustered index, it must be an ordering field in the secondary index. In the Antelope format, only prefix-indexed columns may be stored off-page in the clustered index record. In the Barracuda format, also fully indexed long CHAR or VARCHAR columns may be stored off-page. */ ut_ad(col->ord_part); if (UNIV_LIKELY_NULL(ext)) { /* See if the column is stored externally. */ const byte* buf = row_ext_lookup(ext, col_no, &len); if (UNIV_LIKELY_NULL(buf)) { if (UNIV_UNLIKELY(buf == field_ref_zero)) { return(NULL); } dfield_set_data(dfield, buf, len); } if (ind_field->prefix_len == 0) { /* In the Barracuda format (ROW_FORMAT=DYNAMIC or ROW_FORMAT=COMPRESSED), we can have a secondary index on an entire column that is stored off-page in the clustered index. As this is not a prefix index (prefix_len == 0), include the entire off-page column in the secondary index record. */ continue; } } else if (dfield_is_ext(dfield)) { /* This table is either in Antelope format (ROW_FORMAT=REDUNDANT or ROW_FORMAT=COMPACT) or a purge record where the ordered part of the field is not external. In Antelope, the maximum column prefix index length is 767 bytes, and the clustered index record contains a 768-byte prefix of each off-page column. */ ut_a(len >= BTR_EXTERN_FIELD_REF_SIZE); len -= BTR_EXTERN_FIELD_REF_SIZE; dfield_set_len(dfield, len); } /* If a column prefix index, take only the prefix. */ if (ind_field->prefix_len) { len = dtype_get_at_most_n_mbchars( col->prtype, col->mbminlen, col->mbmaxlen, ind_field->prefix_len, len, dfield_get_data(dfield)); dfield_set_len(dfield, len); } } ut_ad(dtuple_check_typed(entry)); return(entry); }
/*******************************************************************//** Builds from a secondary index record a row reference with which we can search the clustered index record. */ UNIV_INTERN void row_build_row_ref_in_tuple( /*=======================*/ dtuple_t* ref, /*!< in/out: row reference built; see the NOTE below! */ const rec_t* rec, /*!< in: record in the index; NOTE: the data fields in ref will point directly into this record, therefore, the buffer page of this record must be at least s-latched and the latch held as long as the row reference is used! */ const dict_index_t* index, /*!< in: secondary index */ ulint* offsets,/*!< in: rec_get_offsets(rec, index) or NULL */ trx_t* trx) /*!< in: transaction */ { const dict_index_t* clust_index; dfield_t* dfield; const byte* field; ulint len; ulint ref_len; ulint pos; ulint clust_col_prefix_len; ulint i; mem_heap_t* heap = NULL; ulint offsets_[REC_OFFS_NORMAL_SIZE]; rec_offs_init(offsets_); ut_a(ref); ut_a(index); ut_a(rec); ut_ad(!dict_index_is_clust(index)); if (UNIV_UNLIKELY(!index->table)) { fputs("InnoDB: table ", stderr); notfound: ut_print_name(stderr, trx, TRUE, index->table_name); fputs(" for index ", stderr); ut_print_name(stderr, trx, FALSE, index->name); fputs(" not found\n", stderr); ut_error; } clust_index = dict_table_get_first_index(index->table); if (UNIV_UNLIKELY(!clust_index)) { fputs("InnoDB: clust index for table ", stderr); goto notfound; } if (!offsets) { offsets = rec_get_offsets(rec, index, offsets_, ULINT_UNDEFINED, &heap); } else { ut_ad(rec_offs_validate(rec, index, offsets)); } /* Secondary indexes must not contain externally stored columns. */ ut_ad(!rec_offs_any_extern(offsets)); ref_len = dict_index_get_n_unique(clust_index); ut_ad(ref_len == dtuple_get_n_fields(ref)); dict_index_copy_types(ref, clust_index, ref_len); for (i = 0; i < ref_len; i++) { dfield = dtuple_get_nth_field(ref, i); pos = dict_index_get_nth_field_pos(index, clust_index, i); ut_a(pos != ULINT_UNDEFINED); field = rec_get_nth_field(rec, offsets, pos, &len); dfield_set_data(dfield, field, len); /* If the primary key contains a column prefix, then the secondary index may contain a longer prefix of the same column, or the full column, and we must adjust the length accordingly. */ clust_col_prefix_len = dict_index_get_nth_field( clust_index, i)->prefix_len; if (clust_col_prefix_len > 0) { if (len != UNIV_SQL_NULL) { const dtype_t* dtype = dfield_get_type(dfield); dfield_set_len(dfield, dtype_get_at_most_n_mbchars( dtype->prtype, dtype->mbminlen, dtype->mbmaxlen, clust_col_prefix_len, len, (char*) field)); } } } ut_ad(dtuple_check_typed(ref)); if (UNIV_LIKELY_NULL(heap)) { mem_heap_free(heap); } }
/*******************************************************************//** Builds an update vector based on a remaining part of an undo log record. @return remaining part of the record, NULL if an error detected, which means that the record is corrupted */ UNIV_INTERN byte* trx_undo_update_rec_get_update( /*===========================*/ byte* ptr, /*!< in: remaining part in update undo log record, after reading the row reference NOTE that this copy of the undo log record must be preserved as long as the update vector is used, as we do NOT copy the data in the record! */ dict_index_t* index, /*!< in: clustered index */ ulint type, /*!< in: TRX_UNDO_UPD_EXIST_REC, TRX_UNDO_UPD_DEL_REC, or TRX_UNDO_DEL_MARK_REC; in the last case, only trx id and roll ptr fields are added to the update vector */ trx_id_t trx_id, /*!< in: transaction id from this undo record */ roll_ptr_t roll_ptr,/*!< in: roll pointer from this undo record */ ulint info_bits,/*!< in: info bits from this undo record */ trx_t* trx, /*!< in: transaction */ mem_heap_t* heap, /*!< in: memory heap from which the memory needed is allocated */ upd_t** upd) /*!< out, own: update vector */ { upd_field_t* upd_field; upd_t* update; ulint n_fields; byte* buf; ulint i; ut_a(dict_index_is_clust(index)); if (type != TRX_UNDO_DEL_MARK_REC) { ptr = trx_undo_update_rec_get_n_upd_fields(ptr, &n_fields); } else { n_fields = 0; } update = upd_create(n_fields + 2, heap); update->info_bits = info_bits; /* Store first trx id and roll ptr to update vector */ upd_field = upd_get_nth_field(update, n_fields); buf = mem_heap_alloc(heap, DATA_TRX_ID_LEN); trx_write_trx_id(buf, trx_id); upd_field_set_field_no(upd_field, dict_index_get_sys_col_pos(index, DATA_TRX_ID), index, trx); dfield_set_data(&(upd_field->new_val), buf, DATA_TRX_ID_LEN); upd_field = upd_get_nth_field(update, n_fields + 1); buf = mem_heap_alloc(heap, DATA_ROLL_PTR_LEN); trx_write_roll_ptr(buf, roll_ptr); upd_field_set_field_no( upd_field, dict_index_get_sys_col_pos(index, DATA_ROLL_PTR), index, trx); dfield_set_data(&(upd_field->new_val), buf, DATA_ROLL_PTR_LEN); /* Store then the updated ordinary columns to the update vector */ for (i = 0; i < n_fields; i++) { byte* field; ulint len; ulint field_no; ulint orig_len; ptr = trx_undo_update_rec_get_field_no(ptr, &field_no); if (field_no >= dict_index_get_n_fields(index)) { fprintf(stderr, "InnoDB: Error: trying to access" " update undo rec field %lu in ", (ulong) field_no); dict_index_name_print(stderr, trx, index); fprintf(stderr, "\n" "InnoDB: but index has only %lu fields\n" "InnoDB: Submit a detailed bug report" " to http://bugs.mysql.com\n" "InnoDB: Run also CHECK TABLE ", (ulong) dict_index_get_n_fields(index)); ut_print_name(stderr, trx, TRUE, index->table_name); fprintf(stderr, "\n" "InnoDB: n_fields = %lu, i = %lu, ptr %p\n", (ulong) n_fields, (ulong) i, ptr); *upd = NULL; return(NULL); } upd_field = upd_get_nth_field(update, i); upd_field_set_field_no(upd_field, field_no, index, trx); ptr = trx_undo_rec_get_col_val(ptr, &field, &len, &orig_len); upd_field->orig_len = orig_len; if (len == UNIV_SQL_NULL) { dfield_set_null(&upd_field->new_val); } else if (len < UNIV_EXTERN_STORAGE_FIELD) { dfield_set_data(&upd_field->new_val, field, len); } else { len -= UNIV_EXTERN_STORAGE_FIELD; dfield_set_data(&upd_field->new_val, field, len); dfield_set_ext(&upd_field->new_val); } } *upd = update; return(ptr); }
/***********************************************************//** Delete marks or removes a secondary index entry if found. @return DB_SUCCESS, DB_FAIL, or DB_OUT_OF_FILE_SPACE */ static ulint row_undo_mod_del_mark_or_remove_sec_low( /*====================================*/ undo_node_t* node, /*!< in: row undo node */ que_thr_t* thr, /*!< in: query thread */ dict_index_t* index, /*!< in: index */ dtuple_t* entry, /*!< in: index entry */ ulint mode) /*!< in: latch mode BTR_MODIFY_LEAF or BTR_MODIFY_TREE */ { btr_pcur_t pcur; btr_cur_t* btr_cur; ibool success; ibool old_has; ulint err; mtr_t mtr; mtr_t mtr_vers; enum row_search_result search_result; log_free_check(); mtr_start(&mtr); btr_cur = btr_pcur_get_btr_cur(&pcur); ut_ad(mode == BTR_MODIFY_TREE || mode == BTR_MODIFY_LEAF); search_result = row_search_index_entry(index, entry, mode, &pcur, &mtr); switch (UNIV_EXPECT(search_result, ROW_FOUND)) { case ROW_NOT_FOUND: /* In crash recovery, the secondary index record may be missing if the UPDATE did not have time to insert the secondary index records before the crash. When we are undoing that UPDATE in crash recovery, the record may be missing. In normal processing, if an update ends in a deadlock before it has inserted all updated secondary index records, then the undo will not find those records. */ err = DB_SUCCESS; goto func_exit; case ROW_FOUND: break; case ROW_BUFFERED: case ROW_NOT_DELETED_REF: /* These are invalid outcomes, because the mode passed to row_search_index_entry() did not include any of the flags BTR_INSERT, BTR_DELETE, or BTR_DELETE_MARK. */ ut_error; } /* We should remove the index record if no prior version of the row, which cannot be purged yet, requires its existence. If some requires, we should delete mark the record. */ mtr_start(&mtr_vers); success = btr_pcur_restore_position(BTR_SEARCH_LEAF, &(node->pcur), &mtr_vers); ut_a(success); old_has = row_vers_old_has_index_entry(FALSE, btr_pcur_get_rec(&(node->pcur)), &mtr_vers, index, entry); if (old_has) { err = btr_cur_del_mark_set_sec_rec(BTR_NO_LOCKING_FLAG, btr_cur, TRUE, thr, &mtr); ut_ad(err == DB_SUCCESS); } else { /* Remove the index record */ if (mode == BTR_MODIFY_LEAF) { success = btr_cur_optimistic_delete(btr_cur, &mtr); if (success) { err = DB_SUCCESS; } else { err = DB_FAIL; } } else { ut_ad(mode == BTR_MODIFY_TREE); /* No need to distinguish RB_RECOVERY_PURGE here, because we are deleting a secondary index record: the distinction between RB_NORMAL and RB_RECOVERY_PURGE only matters when deleting a record that contains externally stored columns. */ ut_ad(!dict_index_is_clust(index)); btr_cur_pessimistic_delete(&err, FALSE, btr_cur, RB_NORMAL, &mtr); /* The delete operation may fail if we have little file space left: TODO: easiest to crash the database and restart with more file space */ } } btr_pcur_commit_specify_mtr(&(node->pcur), &mtr_vers); func_exit: btr_pcur_close(&pcur); mtr_commit(&mtr); return(err); }